From 28bd88fd90570f3e400de5e54ca7ab2432e3826b Mon Sep 17 00:00:00 2001
From: Florian Pritz <bluewind@xssn.at>
Date: Wed, 29 Sep 2010 20:37:46 +0200
Subject: watchdog.sh: misc changes

Signed-off-by: Florian Pritz <bluewind@xssn.at>
---
 watchdog.sh | 43 +++++++++++++++++++++++++++++--------------
 1 file changed, 29 insertions(+), 14 deletions(-)

(limited to 'watchdog.sh')

diff --git a/watchdog.sh b/watchdog.sh
index 0101b26..ae5e482 100755
--- a/watchdog.sh
+++ b/watchdog.sh
@@ -17,7 +17,7 @@
 #-------------------------- CONFIGURATION -----------------------------#
 # Time to wait between 2 checks; Shouldn't be lower than 10 as that
 # could cause overlap with the timeouts
-WAIT=60
+WAIT=30
 
 # Take action after x faild checks
 MAX_FAIL=2
@@ -38,6 +38,8 @@ EXIT_FAILURE=1
 EXIT_ERROR=2
 EXIT_BUG=10
 
+DEBUG=0
+
 # Colors for output
 red='\e[0;31m'
 RED='\e[1;31m'
@@ -50,16 +52,23 @@ CYAN='\e[1;36m'
 NC='\e[0m'
 
 function usage {
-  echo -e "${blue}Usage:${NC} ${SCRIPTNAME} [OPTIONS] [<hostname or IP> <service type>]" >&2
+  echo -e "${blue}Usage:${NC} ${SCRIPTNAME} [OPTIONS]" >&2
   echo -e "Possible service types: http, ping (default)"
   echo -e "Options:" >&2
   echo -e "-h           this help" >&2
+  echo -e "-d           debug output" >&2
   echo -e "-f <file>    loads IP list from a file" >&2
   echo -e "-t <time>    Time between 2 checks"
   echo -e "-m <tries>   Max. fails before taking actions"
   [[ $# -eq 1 ]] && exit $1 || exit $EXIT_FAILURE
 }
 
+dbg() {
+  if [[ $DEBUG -gt 0 ]]; then
+    echo "$@" >&2
+  fi
+}
+
 if [ ! -f "$CURL" ]; then 
   if [ ! -f "$WGET" ]; then
     echo -e "${red}Wget doesn't exist!$NC"
@@ -81,7 +90,7 @@ if [ ! "$1" ] && [ ! "$2" ]; then
   usage $EXIT_SUCCESS
 fi
 
-while getopts ':f:ht:m:' OPTION ; do
+while getopts ':f:hdt:m:' OPTION ; do
   case $OPTION in
     f)
       IPFILE="$OPTARG"
@@ -95,6 +104,9 @@ while getopts ':f:ht:m:' OPTION ; do
     h)
       usage $EXIT_SUCCESS
 	  ;;
+    d)
+      DEBUG=1
+	  ;;
     \?)
       echo "Unknown option \"-$OPTARG\"." >&2
       usage $EXIT_ERROR
@@ -113,15 +125,12 @@ done
 
 shift $(( OPTIND - 1 ))
 
-if (( $# > 1 )) ; then
-  INPUT="$1 $2"
-else
-  INPUT=$(cat $IPFILE)  
-fi
-
 watcher () {
   IP="$1"
   TYPE="$2"
+  WAIT="$3"
+
+  dbg "watcher: $IP - $TYPE - $WAIT"
   
   case "$TYPE" in
     http) COMMAND="$HTTP";;
@@ -141,12 +150,15 @@ watcher () {
       fi
       let counter=$counter+1
       if [ "$counter" -eq $MAX_FAIL ]; then
-        mail root -s "$IP - $TYPE DOWN! " <<< "$IP - $TYPE is down since $downtime"
+        mail root -s "$IP - $TYPE DOWN! " <<< "$IP - $TYPE is down since $downtime
+$(tracepath $IP)"
+        dbg "$IP - $TYPE down since $downtime"
       fi
     else
       if [ "$counter" -gt $MAX_FAIL ] || [ "$counter" -eq $MAX_FAIL ]; then
         mail root -s "$IP - $TYPE UP! " <<< "$IP - $TYPE is OK again.
 	        Downtime: ${downtime} - $(date)"
+        dbg "$IP - $TYPE up again $(date)"
         echo "$IP - $TYPE Downtime: ${downtime} - $(date)" >> $LOGFILE
       fi
       counter=0
@@ -163,21 +175,24 @@ watcher () {
 
 # stops watchers and exits
 cleanup() {
+  dbg "cleaning up..."
   kill $jobs &> /dev/null
   exit
 }
-trap cleanup 0 2 15
+trap cleanup 2 15
 
 # start watchers
 while read line
 do
   watcher $line & 
-done <<< $INPUT 
+done < $IPFILE
 
 # Prepare killing all watchers
 jobs=$(jobs -p)
 disown -ar
 
-echo "Press any key to kill"
-read -n 1
+echo "Press ^C to kill"
+while :; do
+  sleep 999
+done
 
-- 
cgit v1.2.3-24-g4f1b